| Idan Chen project - Personal Key Indicators of Heart Disease

LinkedIn | Github |

---

| Table Of Contecnts:

• Imports
• Data Preparation
  • Missing values or outliers
  • Couple visualizations
  • Find the best columns for ML on HeartDisease based on their Corr
  • Dummy variables for categorical variables
  • Imblance
• Models
  • Logistic Regression
  • KNN
  • DT
• Comparison
• Summary

---

| Imports

# basic imports
import pandas as pd
import numpy as np
from collections import Counter

# plots
import seaborn as sns
import matplotlib.pyplot as plt
import plotly
import plotly.express as px
import plotly.graph_objects as go
from plotly.subplots import make_subplots

# settings immports
import warnings
warnings.filterwarnings('ignore')
pd.options.display.max_columns = 40
import plotly.io as pio
# notebook == HTML, iframe == Jupyter
#pio.renderers.default = "iframe"
pio.renderers.default = "notebook"
%config Completer.use_jedi = False

# Unbalance data imports
from imblearn import under_sampling, over_sampling
from imblearn.over_sampling import RandomOverSampler
from imblearn.under_sampling import RandomUnderSampler

# ML imports
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsClassifier
from sklearn import tree
from sklearn.ensemble import RandomForestClassifier
from sklearn.tree import plot_tree
import sklearn.linear_model as sklin
import sklearn.metrics as met

---

| Data preperation

df = pd.read_csv('Data/heart_2020_cleaned.csv')

df.head()

	HeartDisease	BMI	Smoking	AlcoholDrinking	Stroke	PhysicalHealth	MentalHealth	DiffWalking	Sex	AgeCategory	Race	Diabetic	PhysicalActivity	GenHealth	SleepTime	Asthma	KidneyDisease	SkinCancer
0	No	16.60	Yes	No	No	3.0	30.0	No	Female	55-59	White	Yes	Yes	Very good	5.0	Yes	No	Yes
1	No	20.34	No	No	Yes	0.0	0.0	No	Female	80 or older	White	No	Yes	Very good	7.0	No	No	No
2	No	26.58	Yes	No	No	20.0	30.0	No	Male	65-69	White	Yes	Yes	Fair	8.0	Yes	No	No
3	No	24.21	No	No	No	0.0	0.0	No	Female	75-79	White	No	No	Good	6.0	No	No	Yes
4	No	23.71	No	No	No	28.0	0.0	Yes	Female	40-44	White	No	Yes	Very good	8.0	No	No	No

---

| Missing values or outliers

df.isna().sum().sum()

0

Tip: zero nan = 😀😀 = make my life much better.

df.nunique()

HeartDisease           2
BMI                 3604
Smoking                2
AlcoholDrinking        2
Stroke                 2
PhysicalHealth        31
MentalHealth          31
DiffWalking            2
Sex                    2
AgeCategory           13
Race                   6
Diabetic               4
PhysicalActivity       2
GenHealth              5
SleepTime             24
Asthma                 2
KidneyDisease          2
SkinCancer             2
dtype: int64

df =  df[df.columns].replace({'Yes':1, 'No':0, 'Male':1,'Female':0,'No, borderline diabetes':'0','Yes (during pregnancy)':'1' })
df['Diabetic'] = df['Diabetic'].astype(int)

---

| Basic visualisations for analysis

I chose to add basic visualisations to show the problem before the model ...

I'll need to use dummy bariables for categorical variables

and other techniques like big imbalance that will destroy my model

histplot of SleepTime devided to 24 bins (number of hours per day)

ax = sns.histplot(data=df,x='SleepTime',bins=24)
_ = ax.set_title("Histribution of Sleep Time With 24 bins")

There are 551 people that sleep 1 hour per day !

And 30 people that sleep 24 hours per day ! ( I think they are died :/)

kdeplot of BMI divided by Heart Disease Label

Explanation: I added BMI photo explanation for checking if the result of the BMI's distribution is ok

fig, ax = plt.subplots(figsize = (13,5))
sns.kdeplot(df[df["HeartDisease"]==1]["BMI"], alpha=0.5,shade = True, color="red", label="HeartDisease", ax = ax)
sns.kdeplot(df[df["HeartDisease"]==0]["BMI"], alpha=0.5,shade = True, color="#fccc79", label="Normal", ax = ax)
plt.title('Distribution of Body Mass Index', fontsize = 18)
ax.set_xlabel("BodyMass")
ax.set_ylabel("Frequency")
ax.legend()
plt.show()

Function that get df and column name and plot countplot with bar labels

def create_count_plot(x_size,y_size, df, column_name):
    fig = plt.figure(figsize=(x_size,y_size))
    plt.title(column_name,fontsize=15)
    ax = sns.countplot(x=df[column_name],
                       order=df[column_name].value_counts(ascending=False).index);
    abs_values = df[column_name].value_counts(ascending=False).values
    _ = ax.bar_label(container=ax.containers[0], labels=abs_values)

count plot of Age Category

create_count_plot(x_size= 14, y_size = 6,df=df, column_name="AgeCategory")

---

| Find the best columns for ML on HeartDisease based on their Corr

fig = px.imshow(df.corr(),color_continuous_scale="Blues")
fig.update_layout(height=600)
fig.show()

plt.figure(figsize = (13,6))
plt.title('Distribution of correlation of features')
abs(df.corr()['HeartDisease']).sort_values()[:-1].plot.barh()
plt.show()

---

After Reaserch by using catplot and pandas corr I Found the suits colums are:

• Smoking
• Stroke
• PhysicalHealth
• DiffWalking
• Sex
• AgeCategory
• Diabetic
• PhysicalActivity
• GenHealth

By the way - It's very intersting to figure out that Sleep Time doesn't effect Heart Disease

Danger: I found the suitests columns but the corrlelation is not great at all (the best is Stroke and DiffWaliking with 0.2 corr)

final_columns = ["Smoking","Stroke","PhysicalHealth","DiffWalking","Sex","AgeCategory","Diabetic","PhysicalActivity","GenHealth","HeartDisease"]

df = df[final_columns]

plt.figure(figsize=(30,30))
for i in enumerate(final_columns[:-1]):
    plt.subplot(6,3,i[0]+1)
    sns.countplot(i[1],hue='HeartDisease',data=df)
    plt.xticks(rotation=15)

---

| Dummy variables for categorical variables

lst_of_dummy_variables = ['HeartDisease','Smoking','Stroke','DiffWalking','Sex','AgeCategory','Diabetic','PhysicalActivity','GenHealth']

df = pd.get_dummies(data=df,columns=lst_of_dummy_variables,drop_first=True)

df.head(1)

	PhysicalHealth	HeartDisease_1	Smoking_1	Stroke_1	DiffWalking_1	Sex_1	AgeCategory_25-29	AgeCategory_30-34	AgeCategory_35-39	AgeCategory_40-44	AgeCategory_45-49	AgeCategory_50-54	AgeCategory_55-59	AgeCategory_60-64	AgeCategory_65-69	AgeCategory_70-74	AgeCategory_75-79	AgeCategory_80 or older	Diabetic_1	PhysicalActivity_1	GenHealth_Fair	GenHealth_Good	GenHealth_Poor	GenHealth_Very good
0	3.0	0	1	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	1	1	0	0	0	1

---

| Balance the data

count plot of Heart Disease

ax= df["HeartDisease_1"].value_counts().plot.pie(autopct='%.2f')
_= ax.set_title("Pie of Heart Disease")

There is 10X more people without Heart Disease then people with Heart Disease :

• we need to do balance with the data - > to lower the number of the people without Heart Disease

we need to fix the dummy varibales and the imbalace in the Y target ('HeartDisease' column)

There are two methods: under sampling and over sampling

I will use under sampling becuse I have a lot of data (27373 rows of Yes target)

and becuse of that I prefer to lower the people without Heart Disease to this number

and not to use Over Sampling to Create more sampling with people with Heart Disease .

If I had small number of people with HeartDisease I'll use Over Sampling

rus = RandomUnderSampler()

X_train, X_test, y_train, y_test = train_test_split(df.drop('HeartDisease_1',axis=1), df['HeartDisease_1'], test_size=0.33, random_state=42)
X_rus_train, y_rus_train = rus.fit_resample(X_train, y_train)
X_rus_test, y_rus_test = rus.fit_resample(X_test, y_test) 

Counter(y_rus_train)

Counter({0: 18225, 1: 18225})

Counter(y_rus_test)

Counter({0: 9148, 1: 9148})

df =  pd.concat([X_rus_train,y_rus_train],axis=1)

Success: We finish to prepere the data and now we can move forward to Model Training

see the data after the Preparation

df.head(1)

	PhysicalHealth	Smoking_1	Stroke_1	DiffWalking_1	Sex_1	AgeCategory_25-29	AgeCategory_30-34	AgeCategory_35-39	AgeCategory_40-44	AgeCategory_45-49	AgeCategory_50-54	AgeCategory_55-59	AgeCategory_60-64	AgeCategory_65-69	AgeCategory_70-74	AgeCategory_75-79	AgeCategory_80 or older	Diabetic_1	PhysicalActivity_1	GenHealth_Fair	GenHealth_Good	GenHealth_Poor	GenHealth_Very good	HeartDisease_1
0	0.0	1	0	0	1	0	0	0	0	0	0	0	0	0	1	0	0	0	1	0	0	0	1	0

---

| Model Training

Important functions for model training

def evaluate_model(model, x_test, y_test):
    """
        This function gets model from skit-learn,
        x_test and y_test type numpy array
        and predict the y result using the model and the x_test.
        Then returns dict with the model result parameters:
        - accuracy
        - precision
        - recall score
        - f1 score
        - roc curve score
        - roc auc score
    """
    # Predict test date 
    y_pred = model.predict(x_test)

    # Calculate basics scores
    acc = met.accuracy_score(y_test, y_pred)
    prec = met.precision_score(y_test, y_pred)
    rec = met.recall_score(y_test, y_pred)
    f1 = met.f1_score(y_test, y_pred)

    # Calculate area under curve (AUC)
    y_pred_proba = model.predict_proba(x_test)[::,1]
    fpr, tpr, _ = met.roc_curve(y_test, y_pred_proba)
    auc = met.roc_auc_score(y_test, y_pred_proba)

    # confussion matrix
    cm = met.confusion_matrix(y_test, y_pred)

    return {'acc': acc, 'prec': prec, 'rec': rec, 'f1': f1 ,
            'fpr': fpr, 'tpr': tpr, 'auc': auc, 'cm': cm}

def print_model_result(model_dict):
    """
        Gets dict of model and prints his numerical results
    """
    print('Accuracy:', model_dict['acc'])
    print('Precision:', model_dict['prec'])
    print('Recall:', model_dict['rec'])
    print('F1 Score:', model_dict['f1'])
    print('Area Under Curve:', model_dict['auc'])
    print('Confusion Matrix:\n', model_dict['cm'])

def draw_cm(title,cm):
    """
        This function get title type str and cm type sklearn.metrics.confution_matrix 
        and draw heatmap of the confution matrix results
    """
    plt.figure(figsize=(7,5))
    ax = sns.heatmap(cm/np.sum(cm),fmt='.2%', annot=True, cmap='Blues')
    ax.set_title(f"Confusion Matrix of {title} Model")
    ax.set_xlabel('\nPredicted Values')
    ax.set_ylabel('Actual Values ');
    ax.xaxis.set_ticklabels(['No HeartDisease','HeartDisease'])
    ax.yaxis.set_ticklabels(['No HeartDisease','HeartDisease'])
    plt.show()

---

| Logistic Regression

model=sklin.LogisticRegression()

model.fit(X_rus_train, y_rus_train)

LogisticRegression()

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

logistic_regression_result = evaluate_model(model, X_rus_test, y_rus_test)

print_model_result(logistic_regression_result)

Accuracy: 0.7608220376038478
Precision: 0.7523799449968267
Recall: 0.7775470048097944
F1 Score: 0.7647564777980862
Area Under Curve: 0.8378536425919472
Confusion Matrix:
 [[6807 2341]
 [2035 7113]]

draw_cm("Logistic Regression",logistic_regression_result["cm"])

---

| DecisionTreeClassifier

model=tree.DecisionTreeClassifier()

model.fit(X_rus_train, y_rus_train)

DecisionTreeClassifier()

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

dt_result = evaluate_model(model, X_rus_test, y_rus_test)

print_model_result(dt_result)

Accuracy: 0.725295146480105
Precision: 0.7310538116591928
Recall: 0.7128334062090075
F1 Score: 0.7218286473323002
Area Under Curve: 0.7654030958427598
Confusion Matrix:
 [[6749 2399]
 [2627 6521]]

draw_cm("DecisionTreeClassifier",dt_result["cm"])

---

| KNeighborsClassifier

model=KNeighborsClassifier()

model.fit(X_rus_train, y_rus_train)

KNeighborsClassifier()

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

knn_result = evaluate_model(model, X_rus_test, y_rus_test)

print_model_result(knn_result)

Accuracy: 0.6364779186707477
Precision: 0.8044379419653743
Recall: 0.36062527328377786
F1 Score: 0.49799984904521094
Area Under Curve: 0.7066569000676626
Confusion Matrix:
 [[8346  802]
 [5849 3299]]

draw_cm("KNeighborsClassifier",knn_result["cm"])

---

| Comparison

# Intitialize figure with two plots
fig, (ax1, ax2) = plt.subplots(1, 2)
fig.suptitle('Model Comparison', fontsize=16, fontweight='bold')
fig.set_figheight(10)
fig.set_figwidth(20)
fig.set_facecolor('white')

# First plot
barWidth = 0.2
logistic_regression_result_score = [logistic_regression_result['acc'], logistic_regression_result['prec'], logistic_regression_result['rec'], logistic_regression_result['f1']]
dt_result_score = [dt_result['acc'], dt_result['prec'], dt_result['rec'], dt_result['f1']]
knn_result_score = [knn_result['acc'], knn_result['prec'], knn_result['rec'], knn_result['f1']]


## Set position of bar on X axis
r1 = np.arange(len(logistic_regression_result_score))
r2 = [x + barWidth for x in r1]
r3 = [x + 2*barWidth for x in r1]

## Make the plot
ax1.bar(r1, logistic_regression_result_score, width=barWidth, edgecolor='white', label='Logistic Regression')
ax1.bar(r2, dt_result_score, width=barWidth, edgecolor='white', label='Decision Tree')
ax1.bar(r3, knn_result_score, width=barWidth, edgecolor='white', label='K-Nearest Neighbors')

## Configure x and y axis
ax1.set_xlabel('Metrics', fontweight='bold')
labels = ['Accuracy', 'Precision', 'Recall', 'F1']
ax1.set_xticks([r + (barWidth * 1.5) for r in range(len(logistic_regression_result_score))], )
ax1.set_xticklabels(labels)
ax1.set_ylabel('Score', fontweight='bold')
ax1.set_ylim(0, 1)

## Create legend & title
ax1.set_title('Evaluation Metrics', fontsize=14, fontweight='bold')
ax1.legend()

# Second plot
## Comparing ROC Curve
ax2.plot(logistic_regression_result['fpr'], logistic_regression_result['tpr'], label='Logistic Regression auc = {:0.3f}'.format(logistic_regression_result['auc']))
ax2.plot(dt_result['fpr'], dt_result['tpr'], label='Decision Tree auc = {:0.3f}'.format(dt_result['auc']))
ax2.plot(knn_result['fpr'], knn_result['tpr'], label='K-Nearest Nieghbor auc = {:0.3f}'.format(knn_result['auc']))

## Configure x and y axis
ax2.set_xlabel('False Positive Rate', fontweight='bold')
ax2.set_ylabel('True Positive Rate', fontweight='bold')

## Create legend & title
ax2.set_title('ROC Curve', fontsize=14, fontweight='bold')
ax2.legend(loc=4)

plt.show()

---

| SUMMARY

When I first started the project I looked for interesting data, I searched on the Kaggle website and find this DataSet: https://www.kaggle.com/datasets/kamilpytlak/personal-key-indicators-of-heart-disease

This data looked interesting to me because it has a lot of comments and likes, and then when I looked into his columns I found a lot of indicators of heart disease: Age, BMI, Storke, Sex and even some indicators I did not even think about them: DiffWalking, kidney disease, Race and exc ...

First of all, I thought about My business needs and I understand what I'm trying to address is to understand how adult people can improve their health status

For the ML my y target is "heart disease" and I will predict the result (0/1) based on my features and when I will know the weight of each feature I will understand how adult people can Improve the chance of not getting heart disease.

After I looked at the data - I started to clean him and delete all the null values, after that, I created basic Visualisations for understanding the data pattern.

When I finished this part I created a heatmap and barh of the corr of the data related to the y-target (heart disease column) and figure out that Diffwalking, Stoke, Diabetic, physical health and kidney disease is the top 5 related to heart disease.

In the next part, I created Dummpy Variables for the ML training part and then Balance the data -> there was the ratio of 1:10 for the non Heart Deasese people so I did Random Under Sample on the data to decrease the people without heart disease and to make the ratio of 1: 1

I used 3 types of ML models:

• Logistic Regression
• KNN
• DT

And then create a plot to see the different on the results between them, I found the

accuracy_score, precision_score, recall_score, f1_score, predict_proba, roc_curve and roc_auc_score

When I look at the final plot that creates the competition between all the scores It seems to me that Logistic Registration is the best fit for this DataSet, It won on every score except for the precision score.

For my Business question: I will say to people that want to decrease their chances to get a heart deasse to:

• start doing physical activity
• Eat healthily
• Check their blood to see if there is a problem related to diabetes or kidney

---

Success: Finish the Heart Disease Project! Thank you very much for reading

# for expor